an overview of ramses - montefiore institute...power system dynamic simulations using a parallel...
Post on 27-Jun-2020
1 Views
Preview:
TRANSCRIPT
An overview of RAMSES(RApid Multiprocessor Simulation of Electric power Systems)
Petros Aristidou and Thierry Van Cutsem
© University of Liège
Dept. Electrical Eng. & Comp. Sc.,
Liège, Belgium
June 2019
Power systems equipped with more and more controls reacting to disturbances, with either beneficial or detrimental effects
requires simulating dynamic responses
static security analysis no longer sufficient
larger and larger models considered
simulation of large interconnections
incorporation of sub-transmission and distribution levels
o distribution grids expected to host more and more distributed energy sources
o active distribution network management impacts overall system response
explicit modelling preferred to equivalents
longer simulated times
check response of system up to its return to steady state
o long-term dynamics : typically several minutes after initiating event
faster than real-time (simulators, prediction capability, etc.)
Dynamic simulation needs
Speeding up simulations
Conventional simulation codes and serial computing platforms have met their limits
dynamic simulation software must be revisited to cope with large models and take advantage of computer technology
multi-core computers available widely and at affordable price
deployed to overcome the limits of single-core computers
parallel computing tools are available
attempts to parallelize existing power system simulation software reveal themselves unsuccessful
new solution schemes must be devised to exploit parallelism
RAMSES was developed with this objective in mind, based on system decomposition
RAMSES ingredients
4
Power system modelling in RAMSES
5
Power system =
AC network(s) + DC grid(s) + injectors + two-ports
Modelling : modularity and flexibility
Each injector :
is interfaced to the AC network through the (rectangular components of) bus voltage and injected current
is modelled with its own Differential-Algebraic (DA) equations
algebraic equations yield higher modelling flexibility
the solver handles equations changing between differential and algebraic
… and similarly for two-ports
Component models (I)
Hard-coded :
synchronous machine, voltage and frequency dependent load
open-source :
induction machine, some IEEE models of excitation and torque controls of synchronous machine, wind generators, etc.
user-defined :
4 categories : torque control of synchronous machine, excitation control of synchronous machine, injector and two-port
compiled and linked to RAMSES for computational efficiency
Component models (II)
Two ways to include models (second currently only under Windows)
Discrete-time controls
9
Monitor some observables from the simulation of the D-A models
modify some parameters in those models
act at discrete times
when a condition is fulfilled, or at multiple of their internal activation period
applied after the simulation time step is completed.
Examples of applications:
distributed controllers
under-frequency and under-voltage load shedding
generator protections, etc.
wide-area monitoring and/or control
tracking state estimation : RAMSES used to simulate SCADA and PMUs
secondary frequency control
secondary voltage control
centralized load shedding against voltage instability
coordinated control of dispersed generation units in distribution grids, etc.
Acceleration techniques – parallel processing
Based on decomposition of model according to :
network(s) + injectors + two-ports
injectors (and two-ports) solved independently of each other
same solution as a non-decomposed scheme by resorting to the Schur-complement for the network equations
tasks pertaining to components assigned to a number of threads, e.g.
update and factorization of local Jacobian
computation of mismatch vector (of Newton method)
computation of contribution to Schur-complement matrix
solution of local linear systems, etc.
threads executed each on a separate processor
computational load balanced among the available processors
shared-memory parallel programming model
through OpenMP Application Programming Interface (API)
Acceleration techniques – localization
After a disturbance, the various components of a (large enough) system exhibit different levels of dynamic activity.
This property is exploited at each time step to :
accelerate Newton scheme
thanks to the decomposed solution scheme, Newton iterations are skipped on components that have already converged
exploit component latency
injectors with high (resp. low) dynamic activity are classified as active; the others as latent
active injectors have their original model simulated
latent injectors are replaced by automatically calculated, sensitivity-based models to accelerate the simulation
a fast to compute metrics is used to classify the injectors
injectors seamlessly switch between categories according to their activity
Acceleration techniques – time-scale decomposition
When only the long-term behaviour of the system is of interest, RAMSES can provide a faster to compute time-averaged response
unlike with the “quasi steady-state” approximation, no model simplification is performed
instead, the original system model is simulated with larger time-steps to “filter out” the fast dynamics
a stiff-decay (or L1-stable) integration scheme is used to this purpose
with a dedicated treatment of the discrete part of the model (limits, switchings, etc.) by the solver.
Example of application:
5-10 seconds after a fault : simulation with time steps of ¼ to ½ cycle
from t 5-10 seconds until t several minutes : simulation with time steps of 2 to 6 cycles
Acceleration techniques
Software implementation
Written in modern (2008) FORTRAN using the OpenMP API for shared-memory parallel programming
general implementation : no “hand-crafted” optimization particular to the computer system, the power system or the disturbance
wide range of platforms : from personal laptops to multi-core scientific computing equipment
Microsoft Windows OS : tested on Windows 7 and 10
Linux OS : tested on Debian, Ubuntu, Redhat
interface with MATLAB
for faster prototyping of discrete-time controls (see slide # 9)
through the “MATLAB engine”
during the simulation RAMSES passes information to MATLAB workspace and receives control actions
Possible execution modes
As a standalone program executed from command line terminal
for remote execution on systems without graphic interfaces
embedded in scripts as part of more complex procedures
with a JAVA-based Graphic User Interface
for an easy-to-use and standalone execution
JAVA for compatibility with multiple platforms
single JAVA archive (.jar file) including all necessary executables and libraries to perform simulations and visualize results software ready to be used with no installation !
a dynamic library (.dll or .so)
to be linked to other software (e.g., written in C)
can be loaded within interpreted languages (e.g., MATLAB or Python)
screenshots of theJAVA-based Graphic User Interface
Example : Hydro-Québec system - 30% motor loads (I)
Characteristics• 2565 buses
• 3225 branches
• 290 power plants
• 4311 dynamically modeled loads
• 506 impedance loads
• 1136 discrete devices
• 35559 differential-algebraic states
Disturbance• short circuit lasting seven cycles
• cleared by opening one line
Simulation• over 240 s
• with one-cycle (16.6 ms) time step
• ϵL=0.1MW/MVAr, Tobs=5 s
Example : Hydro-Québec system - 30% motor loads (II)
Characteristics• same as previous slide
Disturbance• same as previous slide
Simulation• one-cycle time step for the first 15 s
• then 0.05 s for the remaining
• ϵL=0.1MW/MVAr, Tobs=5 sMain speedup: Parallel computing
Main speedup: Localization & time-scale decomposition
Example : PEGASE test system (I)
Characteristics• 15226 buses
• 21765 branches
• 3483 power plants
• 7211 dynamically modeled loads
• 2945 discrete devices
• 146239 differential-algebraic states
Disturbance• busbar fault lasting five cycles
• cleared by opening two double-circuit lines
Simulation• over 240 s
• with one-cycle (20 ms) time step
• ϵL=0.1MW/MVAr, Tobs=20 s
Main speedup: Parallel computing
Main speedup: Localization
Example : PEGASE test system (II)
Characteristics• same as previous slide
Disturbance• same as previous slide
Simulation• over 240 s
• one-cycle time step for the first 15 s
• then 0.05 s for the remaining
• ϵL=0.1MW/MVAr, Tobs=20 s
Main speedup: Parallel computing
Main speedup: Localization & time-scale decomposition
References
D. Fabozzi, A. Chieh, B. Haut, and T. Van Cutsem, “Accelerated and localized newton schemes for faster dynamic simulation of large power systems,” IEEE Trans. on Power Systems, 2013.
P. Aristidou, D. Fabozzi, and T. Van Cutsem, “Dynamic simulation of large-scale power systems using a parallel Schur-complement-based decomposition method,” IEEE Trans. on Parallel and Distributed Systems, 2014.
P. Aristidou, S. Lebeau, T. Van Cutsem. Power System Dynamic Simulations Using a Parallel Two-Level Schur-Complement Decomposition. IEEE Transactions on Power Systems, 2016.
P. Aristidou, S. Lebeau, L. Loud, T. Van Cutsem. Prospects of a new dynamic simulation software for real-time applications on the Hydro-Quebec system. CIGRE Science & Engineering, 2016.
top related